概要

We have developed a new method for matching endoscopic image sequences in order to reconstruct the 3-D structure of the intestines. Endoscopic images are among the most difficult images for identifying correspondences due to the lack of textures. It is difficult not only to detect feature points in the images but also to identify correspondences between the images. As a result, conventional interest point detectors and matching methods can obtain few correct correspondences. In this report, we propose a new feature detection and matching method for endoscopic images. This method focuses on the folds of the intestines because these folds are distinct and stable features in endoscopic images. First, the folds are enhanced in the images. Then, points on the folds are determined. By adopting a HOG-based descriptor for the descriptions of the points, the correspondences between the images can be identified. We …